Demand for Nvidia's H20 chip lackluster

Nvidia's H20 artificial intelligence chip, designed specifically for the Chinese market in compliance with Washington's export controls, is encountering tepid demand, according to industry sources.

In the rapidly developing market, this reluctance opens a strategic window for ambitious Chinese AI chip designers to expand their market share amid growing domestic efforts to hone chip prowess, they added.

"The demand for H20 chips was lower than expected, according to the latest feedback of our customers," said a regional manager of a major Chinese information technology distributor, on condition of anonymity.

The manager said that for chips like the H20, demand depends heavily on large corporations, which are reluctant to buy the products of Nvidia, a United States-based semiconductor company, after the Cyberspace Administration of China summoned the US company for alleged cybersecurity threats in late July.

An engineer at a Chinese internet company, also on condition of anonymity, said: "Cybersecurity is of crucial importance to us. We cannot tolerate any potential risks. Even if H20 chips are available now, we are not that eager to buy."

Pan Helin, a member of the Ministry of Industry and Information Technology's Expert Committee for Information and Communication Economy, said: "If Nvidia's chips really have backdoor risks, that will become its self-dug grave. Users, not only companies in China, but also clients across the world, may abandon their chips over fears of remote shutdowns or data theft."

The Financial Times reported on Monday that Nvidia has agreed to pay the US government 15 percent of its revenue from H20 chip sales in China as part of a deal to secure export licenses to Beijing.

In response to the report, Nvidia said in a statement to China Daily, "We follow rules the US government sets for our participation in worldwide markets."

Charlie Dai, vice-president and principal analyst at market research company Forrester, said the 15 percent fee agreement represents an "unprecedented" move.

"It starkly underscores how trade tensions drive up market access costs, creating substantial financial pressure and strategic ambiguity for technology firms worldwide," Dai said.

Xiang Ligang, director-general of the Zhongguancun Modern Information Consumer Application Industry Technology Alliance, a telecom industry association, said, "The US government's move will fuel Chinese companies' distrust toward Nvidia, which will accelerate the adoption of domestic AI chips."

According to Xiang, Nvidia's dominance in the global AI chip market relies not only on chip performance, but also on its entrenched CUDA ecosystem — a 20-year-old framework for AI compatibility. While Chinese chips such as Huawei's Ascend series can rival the H20 technically, they lack CUDA's universal adaptability for AI large language models.

"However, if security concerns and the 15 percent 'geopolitical premium' drive users toward Chinese chips, developers could shift their focus to compatibility layers for domestic alternatives, eroding Nvidia's ecosystem advantage," he added.

Ecosystem booming

China's domestic AI chip ecosystem is moving quickly. In July, StepFun, a Shanghai-based Chinese AI company, formed a "model-chip synergy ecosystem innovation alliance" with nearly 10 leading domestic AI chip and infrastructure companies, including Huawei, Moore Threads, Beijing Biren Technology Development Co and Cambricon Technology.

The alliance's core mission is clear: to collaboratively optimize large language models to run efficiently on domestic AI chips.

The goal is to create a virtuous cycle. As more AI workloads shift to Chinese chips, economies of scale will drive costs down and improve performance and ecosystem maturity — making homegrown solutions increasingly viable and attractive, according to StepFun.

"For domestic AI chips to better support China's homegrown large language models, the first priority is ensuring a sufficient supply of advanced computing power," said Zhang Wen, founder, chairman and CEO of Beijing Biren Technology Development Co. "The production capacity of advanced domestic AI chips will be crucial for the future of China's AI industry."

masi@chinadaily.com.cn